In partnership with
The document you did not write
Somewhere in your organisation, a document is being drafted right now that you will publish under your name and did not write.
That part is not new. Agencies, consultants, freelancers and law firms have always drafted things for you. What is new is that as of three weeks ago, some of those documents carry an invisible mark recording how they were made, and your contracts almost certainly do not require anyone to tell you which ones.
Until three weeks ago, there was no reason to ask.
The reassurance that answered the wrong question
On 11 August, Anthropic confirmed that Claude now embeds an invisible watermark in the text it generates, and provenance metadata in image files. It applies worldwide to models launched from 2 August onwards. Older models follow by 2 December. OpenAI, Google, Meta, Microsoft and around 190 others have signed the same European code of practice.
The reason is Article 50 of the EU AI Act, which became enforceable on 2 August.
What actually changed on 2 August
The first is technical. AI generated text now carries a mark. Anthropic has been clear that this changes nothing about who owns the output, and that the mark shows only that Claude processed the text, not that Claude wrote it. That is true and worth saying.
The second is legal, and it is the one that matters to you. Article 50 creates a disclosure duty. It does not sit with Anthropic. It sits with the organisation publishing the content.
So when the reassurance says nothing changes, it is describing the mark. It is not describing the law. Ownership was never the risk. Disclosure is.
The exemption is the whole game
Article 50 requires organisations to disclose AI generated text published to inform the public on matters of public interest. Commission guidance points to economic, environmental, political and scientific questions of public debate, and to corporate reporting. In practice that means the sustainability report, a public consultation response, an investor communication. Routine marketing copy is generally outside it.
The duty falls away in one situation: where a person has genuinely reviewed the content and holds editorial responsibility for publishing it.
That exemption is the whole game. It is not something you can simply assert. It is something you would need to show.
Why the watermark will not save you
Two problems, briefly.
It cannot tell the difference between a document Claude wrote from scratch and one where Claude fixed the grammar. Both look the same to a detector. Yet that is precisely the distinction the law treats as decisive.
And it weakens as text is edited. A lightly tidied AI draft usually keeps the mark. A heavily rewritten one usually loses it. So detection measures how much editing happened, not how much thought went in. Used as a test of honesty or quality, it catches the careful and misses the careless.
Anthropic announced a detection tool for third parties on 14 August, which is welcome. But a detector helps whoever is making the accusation. It does not help you answer one.
Six questions for your next board meeting
This is not an AI policy question. Most organisations already have a policy covering which tools staff may use. This is a disclosure and records question. Six items belong on the agenda.
Scope. Does any of our published output reach the European Union? If yes, Article 50 may apply to us directly. If no, we still receive the technical marking, but the legal duty may not bite. Have this determined, not assumed.
The in scope list. Which specific publications count as informing the public on matters of public interest? Name them. For most organisations it is a short list: annual and sustainability reports, regulatory and consultation submissions, investor material, public position statements. Most other output sits outside.
Named editorial responsibility. For each item on that list, who holds editorial responsibility? A name, recorded. Not a department.
Evidence of review. What does meaningful human review mean here, and could we demonstrate it happened, eighteen months later, to someone unsympathetic? If the honest answer is a shrug, that is the gap.
The supply chain. The document I opened with, and the one most boards will miss. Agencies, consultants, freelancers and law firms are producing content for you using these tools right now. Do your contracts require them to disclose AI use to you? Almost certainly not, because until three weeks ago there was no reason to ask. Your exposure arrives through work you did not produce and cannot inspect.
Response readiness. If a client, journalist or opposing party runs a detector across our material and makes an allegation, who responds, and with what? Deciding that in advance costs an hour. Deciding it under pressure costs a great deal more.
Two things to watch
The 2 December deadline, when the remaining older models come into scope. And whether marking extends to audio and video, which it currently does not, and which is where the harder problems sit.
The point
Nobody is going to ask whether you own your content. They are going to ask how it was made. The organisations that struggle here will not be the ones using AI. They will be the ones that never decided what to say about it, and cannot show who was responsible.
In July I asked who answers for the version of you an AI speaks aloud. This is the same question from the other side: not what the machine says about you, but what you are obliged to say about the machine.
For transparency: this edition was drafted with AI assistance, checked against the sources below, and published under my editorial responsibility. It is almost certainly carrying the mark it describes.
The Signal
Anthropic's watermark changes nothing about who owns your content. Article 50 of the EU AI Act changes who has to disclose how it was made. The first is a fact about a piece of technology. The second is an obligation on your organisation. They are not the same thing.
The Question
Take the last three documents your organisation published on a matter of public interest. Can you name the person who held editorial responsibility for each, and could you evidence the review they carried out? If not, the work here is not technical. It is a records problem, and it is fixable this quarter.
Sources and Further Reading
Anthropic, How Claude marks AI generated content, Claude Help Center: https://support.claude.com/en/articles/16266773-how-claude-marks-ai-generated-content
TechCrunch, Anthropic says it will watermark text generated by its AI models, 11 August 2026: https://techcrunch.com/2026/08/11/anthropic-says-it-will-watermark-text-generated-by-its-ai-models/
Axios, Anthropic's text watermarks signal new front in AI detection, 12 August 2026: https://www.axios.com/2026/08/12/anthropic-claude-watermarks-ai-detection
Artificial Lawyer, Anthropic will embed watermarks in AI outputs, 13 August 2026: https://www.artificiallawyer.com/2026/08/13/anthropic-will-embed-watermarks-in-ai-outputs/
The Decoder, Anthropic announces watermark detection API for third parties, 14 August 2026: https://the-decoder.com/anthropic-announces-watermark-detection-api-that-will-let-third-parties-detect-claudes-ai-texts/
European Commission, Transparency obligations under Article 50 of the AI Act, questions and answers: https://digital-strategy.ec.europa.eu/en/faqs/transparency-obligations-under-article-50-ai-act
EU Artificial Intelligence Act, Article 50: https://artificialintelligenceact.eu/article/50/
William Fry, A practical guide to the extraterritorial reach of the AI Act: https://www.williamfry.com/knowledge/a-practical-guide-to-the-extraterritorial-reach-of-the-ai-act/
The views and analysis in this newsletter are Jonathan Ashton's own opinion, based on the facts set out above and on the published sources listed. They are not legal advice, and the summary of Article 50 here is a simplification. Organisations whose output reaches the European Union should take their own advice on whether and how these obligations apply to them.
This edition was researched and drafted with AI assistance, checked against the primary sources listed above, edited by me, and published under my editorial responsibility.
If this was forwarded to you, or you found it on LinkedIn: the full edition, with The Signal and the sources, goes out by email first. Subscribe to Common Ground.
The best marketing ideas come from marketers who live it.
That’s what this newsletter delivers.
The Marketing Millennials is a look inside what’s working right now for other marketers. No theory. No fluff. Just real insights and ideas you can actually use—from marketers who’ve been there, done that, and are sharing the playbook.
Every newsletter is written by Daniel Murray, a marketer obsessed with what goes into great marketing. Expect fresh takes, hot topics, and the kind of stuff you’ll want to steal for your next campaign.
Because marketing shouldn’t feel like guesswork. And you shouldn’t have to dig for the good stuff.
Sponsor Common Ground
Common Ground reaches around 1,500 senior legal, governance, risk and communications professionals across the GCC, UK and wider MENA region. One sponsor per issue, clearly labelled. See rates and availability.
Work with me
I advise boards and executive teams on reputation, disclosure and communications risk. Get in touch.